"code": "connection.execute(\" CREATE OR REPLACE FUNCTION timestamp_id(table_name text)\\n RETURNS bigint AS\\n $$\\n DECLARE\\n time_part bigint;\\n sequence_base bigint;\\n tail bigint;\\n BEGIN\\n time_part := (\\n -- Get the time in milliseconds\\n ((date_part('epoch', now()) * 1000))::bigint\\n -- And shift it over two bytes\\n << 16);\\n\\n sequence_base := (\\n 'x' ||\\n -- Take the first two bytes (four hex characters)\\n substr(\\n -- Of the MD5 hash of the data we documented\\n md5(table_name ||\\n '#{SecureRandom.hex(16)}' ||\\n time_part::text\\n ),\\n 1, 4\\n )\\n -- And turn it into a bigint\\n )::bit(16)::bigint;\\n\\n -- Finally, add our sequence number to our base, and chop\\n -- it to the last two bytes\\n tail := (\\n (sequence_base + nextval(table_name || '_id_seq'))\\n & 65535);\\n\\n -- Return the time part and the sequence part. OR appears\\n -- faster here than addition, but they're equivalent:\\n -- time_part has no trailing two bytes, and tail is only\\n -- the last two bytes.\\n RETURN time_part | tail;\\n END\\n $$ LANGUAGE plpgsql VOLATILE;\\n\")",
"code": "ActiveRecord::Base.connection.execute(\" CREATE OR REPLACE FUNCTION timestamp_id(table_name text)\\n RETURNS bigint AS\\n $$\\n DECLARE\\n time_part bigint;\\n sequence_base bigint;\\n tail bigint;\\n BEGIN\\n -- Our ID will be composed of the following:\\n -- 6 bytes (48 bits) of millisecond-level timestamp\\n -- 2 bytes (16 bits) of sequence data\\n\\n -- The 'sequence data' is intended to be unique within a\\n -- given millisecond, yet obscure the 'serial number' of\\n -- this row.\\n\\n -- To do this, we hash the following data:\\n -- * Table name (if provided, skipped if not)\\n -- * Secret salt (should not be guessable)\\n -- * Timestamp (again, millisecond-level granularity)\\n\\n -- We then take the first two bytes of that value, and add\\n -- the lowest two bytes of the table ID sequence number\\n -- (`table_name`_id_seq). This means that even if we insert\\n -- two rows at the same millisecond, they will have\\n -- distinct 'sequence data' portions.\\n\\n -- If this happens, and an attacker can see both such IDs,\\n -- they can determine which of the two entries was inserted\\n -- first, but not the total number of entries in the table\\n -- (even mod 2**16).\\n\\n -- The table name is included in the hash to ensure that\\n -- different tables derive separate sequence bases so rows\\n -- inserted in the same millisecond in different tables do\\n -- not reveal the table ID sequence number for one another.\\n\\n -- The secret salt is included in the hash to ensure that\\n -- external users cannot derive the sequence base given the\\n -- timestamp and table name, which would allow them to\\n -- compute the table ID sequence number.\\n\\n time_part := (\\n -- Get the time in milliseconds\\n ((date_part('epoch', now()) * 1000))::bigint\\n -- And shift it over two bytes\\n << 16);\\n\\n sequence_base := (\\n 'x' ||\\n -- Take the first two bytes (four hex characters)\\n substr(\\n -- Of the MD5 hash of the data we documented\\n md5(table_name ||\\n '#{SecureRandom.hex(16)}' ||\\n time_part::text\\n ),\\n 1, 4\\n )\\n -- And turn it into a bigint\\n )::bit(16)::bigint;\\n\\n -- Finally, add our sequence number to our base, and chop\\n -- it to the last two bytes\\n tail := (\\n (sequence_base + nextval(table_name || '_id_seq'))\\n & 65535);\\n\\n -- Return the time part and the sequence part. OR appears\\n -- faster here than addition, but they're equivalent:\\n -- time_part has no trailing two bytes, and tail is only\\n -- the last two bytes.\\n RETURN time_part | tail;\\n END\\n $$ LANGUAGE plpgsql VOLATILE;\\n\")",
Eugen Rochko
6 years ago
GitHub